Terminal‑Bench 2.1 shows Claude Code + Fable 5 topping the leaderboard with an 83.8% success rate, revealing that raw model strength alone does not guarantee terminal performance and that the engineering of the Agent Harness is the decisive factor.
SpaceXAI's Grok 4.5 arrives ahead of OpenAI's GPT‑5.6 Sol, delivering terminal‑bench scores of 83.3% and multilingual code‑fixing at 78.0%, while offering a striking $2/$6 per‑million‑token price and 80 TPS inference speed, yet still lagging behind top rivals on the hardest SWE‑Bench Pro tasks.
OpenAI unveiled three GPT‑5.6 models—Sol, Terra and Luna—featuring tiered pricing, record‑breaking benchmark scores in programming, security and biology, new max and ultra inference modes, and a limited early rollout, while also noting several unexpected failure cases.
The author compares Claude 4.8 (Opus) and Codex 5.5 across SWE‑bench Pro (69.2% vs 58.6%) and Terminal‑Bench (78.2% vs 74.6%), highlighting Claude’s larger 1 M‑token context, higher accuracy on complex multi‑file tasks, and higher cost, while Codex offers faster, cheaper terminal‑focused performance, recommending each for specific scenarios.
Stanford, Berkeley and Nvidia researchers introduce LLM-as-a-Verifier, a universal verification framework that enhances agent performance, safety and stability on long‑horizon tasks, and outperforms Claude Mythos and GPT‑5.5 on the Terminal‑Bench and SWE‑Bench benchmarks.
The paper introduces Agentic Harness Engineering (AHE), an observability‑driven framework that automatically evolves coding‑agent harnesses, boosting GPT‑5.4's pass@1 score on Terminal‑Bench 2 from 69.7% to 77.0% (+7.3 points), achieving a worldwide top‑three ranking and demonstrating strong cross‑task and cross‑model generalization.
This article walks through ten mainstream open‑source large‑model benchmarks—SWE‑bench Verified and Pro, MMLU‑Pro, GPQA Diamond, HLE, AIME, HMMT, olmOCR‑bench, Terminal‑Bench 2.0, and EvasionBench—explaining their data, evaluation metrics, current leading models, and the capability dimensions they reveal.
Stanford, Berkeley and Nvidia introduce LLM‑as‑a‑Verifier, a verification framework that scales verification compute, uses fine‑grained score tokens, repeated checks and criteria decomposition to boost agent performance, eliminate scoring ties and achieve SOTA results on Terminal‑Bench, surpassing Claude Mythos and GPT‑5.5 while improving safety in long‑horizon tasks.
OpenAI’s April 23, 2026 launch of GPT-5.5 delivers record‑high scores on SWE‑Bench Pro (58.6%) and Terminal‑Bench 2.0 (82.7%), adds persistent multi‑file context, dynamic reasoning time, and token efficiency, while real‑world case studies show substantial productivity gains across engineering teams.
In a landscape where Claude Code, Cursor and Windsurf pile on built‑in features, the pi terminal coding agent adopts a minimalist "primitives, not features" philosophy, removes most defaults, offers extensible CLI tools, supports dozens of LLM providers, and outperforms its rivals in Terminal‑Bench 2.0.
